TensorFlow implementation of the deeply supervised attention-gated residual U-Net trained to detect extrusion events in time-lapse brightfield images of a cell monolayer.
The quantification of extrusion events in brightfield images is a crucial component in our understanding of cellular dynamics. However, manually counting these events can be a laborious process. Thus, we sought to use a machine learning approach to detect these extrusions. A deeply supervised attention-gated residual U-Net was trained using Tensorflow and the output of the network is shown in the image below. Please see our paper for full details of the implementation.
The following packages are required. Please ensure that they are installed using either pip or conda.
- tensorflow-gpu
- tensorpack
- albumentations
- tqdm
- numpy
- scipy
- scikit-image
Please ensure that the required packages are installed.
The training data consist of images consist of 5 consecutive time frames of size 128x128px. These 5 time frames can be broken down to the two time point before the frame of interest, the frame of interest itself, and the two time points after the frame of interest. The ground truth is a binary mask of size 128x128px with the extrusion events given a value of 1.
All the parameters can be set up by modifying the respective variables in default_unet.yml. This will then be loaded during the initialization of the network. An additional copy is saved in the folder of the trained network to facilitate the recording of parameters used.
Please see Training.ipynb for an example of how to train the network. Please see Prediction.ipynb for an example of how to run the prediction model on the dataset.